Name: Jan Ciesko
Pronouns: he
Biography:
Jan Ciesko is a Senior Member of Technical Staff at Sandia National Laboratories in the scalable algorithms group. He holds a master’s degree in computer science from Friedrich-Alexander Universität in Germany and a Ph.D. from Universitat Politècnica de Catalunya (BarcelonaTech) in Spain. Jan works on programmability and performance portability of software on modern supercomputers. He is the developer of PGAS support in the Kokkos Performance Portability Framework and leads development of user-level threading libraries at Sandia. He periodically contributes to Kokkos, MPICH, Libevent, Open MPI and other software packages. He participates in industry forums such as NVIDIA Corporation’s Summit on Summit, Sierra, and Perlmutter (SoSSP), and the Open MPI Forum and has developed collaborations with industry and academic partners including AMD Research, HPE, GeorgiaTech, and the University of Tennessee (UTK) to cover Sandia’s research interests in his field. Before joining the Sandia National Laboratories, he was a post-doc at the Barcelona Supercomputing Center. His work has been published at IPDPS, ICS, HPEC, IWOMP and ExaMPI.
Institution/Lab: Sandia National Laboratories
Website: https://www.sandia.gov/ccr/staff/jan-ciesko-clone/
SRP Collaboration Topic/Title: Next Gen Communication APIs in HPC
Field or research area: GPU Parallel Programming
Please select all the topical areas that apply to your project:
Computer Science (i.e., architectures, compilers/languages, networks, workflow/edge, experiment automation, containers, neuromorphic computing, programming models, operating systems, sustainable software); High-Performance Computing
Brief Abstract:
Communication offloading to GPUs represents a hot topic in science and industry. While many algorithms could benefit from this capability in terms of performance and programmability, novel communication APIs are required that are more suitable for GPU execution. In this work we examine the current PGAS support for distributed parallel programming in Kokkos and evaluate that implementation in comparison to proposals coming from the MPI Forum as well as hardware manufactures such as NVIDIA. Using micro-benchmarks we showcase performance and programmability differences and conclude this work with recommendations on future communication APIs for scientific computing and HPC.
Desired relevant skills, background, or interests:
Parallel Programming, GPU Programming, C++
Other comments:
Do any special requirements apply? In-Person Only; U.S. Citizen Only; Permanent Resident OK; International OK
Other, specify:
Keywords:
MPI, GPU, PGAS, HPC
Lightning Talk Title: The Age of Supernodes